A real-time data processing method and system for multi-party risk labels

Abstract

The present invention discloses a multi-party risk label real-time data processing method and system, the method comprising the following steps: the data party obtains local user risk label data, and generates an aligned data set through label alignment processing for secret sharing or homomorphic encryption processing; the task participant obtains encrypted data, performs joint training based on the encrypted data, generates a basic model, and deploys the trained basic model to the business system; the business system continuously collects new user risk label data, performs label alignment processing on the new data, and then performs secret state processing through secret sharing or homomorphic encryption; performs joint training based on secret state incremental data to generate an incremental model; combines the basic model with the incremental model for model fusion, generates a fusion model to replace the basic model and deploy it to the business system. The present invention can improve the computational efficiency of multi-party risk label real-time data processing and meet the business system's requirements for low latency.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a multi-party risk label real-time data processing method and system. Background Art

[0002] Telecom fraud is a new type of crime that seriously endangers the property safety of citizens. With the continuous advancement of communication technology, telecom fraud cases are emerging in an endless stream around the world, with various means, making it difficult to prevent. In order to accurately identify the risk of telecom fraud, the public security organs have joined forces with financial institutions and telecom operators (also known as Internet Service Providers, ISPs) to build a joint anti-fraud model, which effectively prevents the occurrence of fraud.

[0003] A Chinese invention patent application with publication number CN115271930A discloses a method and device for identifying abnormal users, which method includes the following steps: obtaining first data of a user on a bank side and second data of a user on an operator side, the first data including first identification information and a first feature, and the second data including second identification information and a second feature; matching and aligning the first data and the second data based on the first identification information and the second identification information; obtaining first parameters of a first neural network model on a bank side and second parameters of a second neural network model on an operator side; using the first parameters and the second parameters to generate a neural network joint model; and using the neural network joint model to identify whether the logged-in user on the bank side is an abnormal user.

[0004] Since the data of public security organs, financial institutions and telecom operators are extremely private and sensitive, the data that can be used to jointly build models is very limited. The above solution is insufficient in data security and privacy protection. This method may involve the direct exposure of user identity information and has a high risk of privacy leakage.

[0005] Multi-party computation (MPC) is a cryptographic technology that allows multiple parties to jointly compute a function and obtain results without leaking their private data. Its core goal is to ensure data privacy and computational correctness, even if some parties behave improperly or do not trust other parties. Therefore, combining privacy computing technology to jointly train an anti-fraud model with the data of multiple parties without leaving the domain is a very worthwhile direction to explore.

[0006] In view of this, a Chinese invention patent application with publication number CN117195060A discloses a telecommunications fraud identification method and model training method based on multi-party secure computing, the method comprising the following steps: obtaining telecommunications data to be identified; using a secret sharing method with a second participant and a third participant, using a target recognition model to process the telecommunications data to be identified, the bank flow data to be identified in the second participant, and the web browsing behavior data to be identified in the third participant, to obtain a first fragment of secret sharing of a target prediction value; receiving a second fragment and a third fragment of secret sharing of a target prediction value; determining a telecommunications fraud identification result based on the first, second, and third fragments; the target recognition model is obtained by combining multiple single tree models; the single tree model is constructed based on fragments of secret sharing of the first-order derivative and the second-order derivative of the target loss function; the target loss function includes a penalty parameter to suppress the imbalance in the number of positive and negative samples.

[0007] In anti-fraud scenarios, it is of key significance to detect possible fraudulent behaviors in a timely manner and quickly implement blocking. The success rate of blocking fraudulent behaviors is highly correlated with the reaction time. The faster the response speed, the higher the probability of successful blocking. Therefore, in response to this demand, the technical solution needs to have the ability of real-time dynamic learning and reasoning, and can also continuously optimize the model to cope with complex and changeable fraud methods. The above-mentioned multi-party secure computing technology has significant problems in model training, including low computing efficiency, high communication overhead, poor scalability, and inability to meet low latency requirements. Especially in the scenario of large-scale data and high-dimensional features, frequent cryptographic operations and synchronization of intermediate results greatly increase the computing and communication costs, and its optimization for specific tasks is still relatively lacking. In addition, solving the trade-off between efficiency and security is still a key difficulty, which makes it difficult to meet industrial-level performance requirements in practical applications. These problems make the large-scale implementation of multi-party secure computing in model training face huge obstacles, and it is urgent to make breakthroughs in algorithm optimization, framework design and system implementation. Summary of the invention

[0008] The present invention provides a real-time data processing method and system for multi-party risk labels, aiming to solve the problems existing in the prior art in the model training of multi-party secure computing, such as low computing efficiency, high communication overhead, poor scalability, and inability to meet low latency requirements.

[0009] In order to solve the above technical problems, the present invention proposes a multi-party risk label real-time data processing method, which includes the following steps: Multiple data providers obtain local user risk label data, and generate aligned data sets through label alignment, and perform secret sharing or homomorphic encryption on the aligned data sets; The task participants obtain the encrypted data, perform joint training based on the encrypted data, generate a basic model, and deploy the trained basic model to the business system; Continuously collect new user risk label data through the business system, perform label alignment on the new data, and then perform confidential processing through secret sharing or homomorphic encryption; Perform joint training based on dense incremental data to generate incremental models; Combine the basic model with the incremental model to perform model fusion, generate a fusion model to replace the basic model and deploy it to the business system.

[0010] Preferably, the label alignment process is as follows: each participant broadcasts its own local label category to other participants for label alignment, and compares the label category received by the party with the locally stored label category, and fills in the missing label category in the local storage with preset data.

[0011] Preferably, the secret sharing is specifically: There are N data providers, each of which holds a local user label dataset; For each user label data , construct a random polynomial whose order is less than the set threshold T:

[0012] In the formula, , indicating that the user On the Participants The The number of tags, is a randomly selected coefficient, T is the minimum number of shares required to reconstruct the data, satisfying ; calculate The value at N different points gives the secret share:

[0013] Each participant Assign a secret share .

[0014] Preferably, the homomorphic encryption is specifically: Select a security parameter and have a trusted party generate a public-private key pair , public key Used to encrypt data, private key Used to decrypt data; Each participant Perform homomorphic encryption on local user label data to generate encrypted label data.

[0015] Preferably, the homomorphic encryption adopts one of the Paillier algorithm, the ElGamal algorithm and the Boneh-Goh-Nissim algorithm.

[0016] Preferably, the model fusion method is specifically as follows: Process the incremental data to make its feature distribution consistent with the basic model input, and generate prediction results for the basic model and incremental model; Use the basic model as the teacher model and the incremental model as the student model to train a fusion model; Calculate the validation set performance of the basic model and the incremental model, assign weights to the basic model and the incremental model according to the validation set performance, and use the weighted average method to fuse the outputs of the basic model and the incremental model.

[0017] Preferably, when training a fusion model, the loss function for:

[0018] In the formula, is the temperature constant used to control the degree of softening, represents the Softmax function, The prediction results of the basic model are is the prediction result of the incremental model, express and The relative entropy of .

[0019] Preferably, in the weighted average method of model fusion, the basic model weights and the incremental model weights are dynamically adjusted through Bayesian optimization.

[0020] Preferably, the joint training adopts a federated learning architecture, where each participant uses local data to train a local model and updates the global model through a secure aggregation mechanism.

[0021] Accordingly, the present invention also proposes a multi-party risk tag real-time data processing system, which is used to implement the above-mentioned real-time data processing method, including: The data acquisition module is used to obtain local user risk label data, continuously collect new user risk label data, and perform label alignment processing to generate an aligned data set; Data encryption module, used to perform secret sharing or homomorphic encryption on the aligned data set; A joint training module, used to obtain encrypted data, and perform joint training based on the encrypted data to generate a basic model; The model deployment module is used to deploy the trained basic model to the business system and apply it in the business system; The incremental training module is used to perform joint training based on dense incremental data to generate incremental models; The model fusion module is used to combine the basic model and the incremental model to generate a fusion model and dynamically adjust the weights according to the validation set performance of the basic model and the incremental model. The model update module is used to replace the basic model with the fusion model and deploy it to the business system.

[0022] Compared with the prior art, the present invention has the following technical effects: 1. The multi-party risk label real-time data processing method proposed in the present invention splits the existing model training process into two stages, namely the basic model training stage and the incremental model training stage. The basic model stage is used with the largest data volume and the longest training time, and this stage is not sensitive to the computational time. In the incremental learning stage, incremental data with a small amount of data is used to train the incremental model. This process greatly reduces the amount of communication in the multi-party secure computing protocol interaction process due to the very small amount of data, speeds up the model training time, and can greatly reduce the computational time, thereby meeting the business system's demand for low-latency reasoning in anti-fraud scenarios.

[0023] 2. The multi-party risk label real-time data processing method proposed in the present invention integrates the basic model with the incremental model through the model fusion method, effectively utilizes historical knowledge, and absorbs the latest information at the same time to avoid the problem of rapid degradation of the model due to changes in data distribution.

[0024] 3. The real-time data processing method for multi-party risk labels proposed in the present invention uses secret sharing and homomorphic encryption technology to encrypt the risk label data of multiple users, realize joint modeling without leaking the original data, and avoid the privacy leakage risk caused by direct data sharing.

[0025] 4. The multi-party risk label real-time data processing method proposed in the present invention supports data collaboration among multiple institutions such as banks and operators through multi-party secure computing. Under the premise of ensuring that the data does not leave the local area and is not directly shared, it realizes cross-institutional joint modeling and improves the accuracy of risk identification. It is suitable for multiple fields such as financial risk control, telecommunications fraud detection, network security monitoring, etc., meets data security compliance requirements, and at the same time improves the risk prediction ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the real-time data processing method of the present invention; Figure 2 This is a schematic diagram of label alignment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0028] Embodiment 1 This embodiment is a real-time data processing method for multi-party risk labels. Figure 1 As shown, the steps include steps one to five: In step 1, multiple data providers obtain local user risk label data, generate an aligned data set through label alignment, and perform secret sharing or homomorphic encryption on the aligned data set.

[0029] The label alignment process is specifically as follows: each participant broadcasts its own local label categories to other participants for label alignment, and compares the label categories received by the party with the locally stored label categories, and fills in the missing label categories in the local storage with preset data.

[0030] In this embodiment, the preset data for filling is the constant 0, that is, the constant 0 is used to fill the labels that the party does not have, so as to achieve label alignment.

[0031] An example of label alignment in this embodiment is as follows Figure 2 As shown in the figure, assume that there are three participants: P1, P2, and P3, each of which holds local user risk label data: P1 has label A and label C, P2 has label A and label B, and P3 has label A and label B, where labels A, B, and C are labels generated by each data party for the corresponding user according to their own business. Each participant broadcasts the local label category to other participants. As shown in the figure, P1 sends A and C to P2 and P3, P2 sends A and B to P1 and P3, and P3 sends A and B to P1 and P2. After receiving the labels from other participants, each party fills the missing labels with 0, that is, there is no user data with this label, and the default value is set to 0. After alignment, the labels of each data party include A, B, and C, where P1's label B=0, P2's label C=0, and P3's label C=0.

[0032] The secret sharing is specifically: There are N data providers, each of which holds a local user label dataset; For each user label data , construct a random polynomial whose order is less than the set threshold T:

[0033] In the formula, , indicating that the user On the Participants The The number of tags, is a randomly selected coefficient, T is the minimum number of shares required to reconstruct the data, satisfying ; calculate The value at N different points gives the secret share:

[0034] Each participant Assign a secret share .

[0035] The homomorphic encryption is specifically: Select a security parameter and have a trusted party generate a public-private key pair , public key Used to encrypt data, private key Used to decrypt data; Each participant Perform homomorphic encryption on local user label data to generate encrypted label data.

[0036] The homomorphic encryption adopts one of the Paillier algorithm, the ElGamal algorithm and the Boneh-Goh-Nissim algorithm.

[0037] After label alignment and secret sharing / homomorphic encryption, the data of all participants has been standardized into the same label structure and encrypted to ensure data privacy. The next step will be to conduct joint training based on these encrypted data to generate a basic model. By using secret sharing and homomorphic encryption technology to encrypt the risk label data of multiple users, joint modeling can be achieved without leaking the original data, avoiding the privacy leakage risk caused by direct data sharing.

[0038] Step 2: The task participants obtain the encrypted data, perform joint training based on the encrypted data, generate a basic model, and deploy the trained basic model to the business system.

[0039] In the previous step, all data providers (participants) completed label alignment and protected the data using secret sharing or homomorphic encryption. The goal of this step is to build a basic model through joint training without leaking the original data and deploy it to the business system.

[0040] like Figure 2 For the label alignment task shown, an example participant of this embodiment may be the original data provider (P1, P2, P3), who can directly use his own encrypted data for training; or a third-party computing service provider such as a trusted computing party, cloud server, etc., who does not own the data but is responsible for secure computing.

[0041] If secret sharing is used, the share data provided by each party is jointly calculated by one or more computing parties; if homomorphic encryption is used, the computing parties can directly perform calculations on the encrypted data without decryption.

[0042] For secret sharing, one embodiment of the present invention adopts a secure computing solution based on secret sharing, where each participant holds a secret share of the data, but cannot restore the original data alone. During training, each computing step is jointly calculated by multiple participants, and the calculation results are finally combined.

[0043] The combined calculation results can be obtained by secret sharing addition:

[0044] In the formula, Indicates user On the Participants The The number of tags, , , represents the encrypted shares provided by the 1st, 2nd, ..., Nth participants, i.e., the 1st, 2nd, ..., Nth participants After the calculation is completed, the gradients are merged through secret reconstruction and the model parameters are updated.

[0045] For homomorphic encryption, one embodiment of the present invention uses additive homomorphism to accumulate gradients. After the calculation is completed, only the authorized party can use the private key to decrypt the gradient and update the model parameters through the gradient descent algorithm (such as SGD).

[0046] In some other embodiments of the present invention, the joint training may also adopt a federated learning architecture, where each participant uses local data to train a local model and updates the global model through a secure aggregation mechanism. Specifically, each participant trains a sub-model locally, then encrypts the gradient parameters and uploads them to an aggregation server; the server performs encrypted aggregation, and the calculated model parameters are then distributed back to each participant.

[0047] At this point, the model has been trained and can be deployed locally for reasoning. However, there is a problem. Due to the bottleneck of communication efficiency in the MPC protocol itself, the training efficiency of large-scale data sets is very poor, which is obviously not suitable for the low latency requirements of anti-fraud scenarios. Therefore, continue to perform the following steps 3 to 5.

[0048] Step three: Continuously collect new user risk label data through the business system, perform label alignment on the new data, and then perform confidential processing through secret sharing or homomorphic encryption.

[0049] As the business runs, new user data is constantly generated. For example, financial institutions, telecom operators, public security agencies and other parties may generate new user risk label data regularly or in real time. At this time, the business system needs to continuously collect these new data from each participant. Each participant regularly processes the new user data and sends it to the central system or computing platform for processing; the new data includes user risk label data, and each label corresponds to specific risk information.

[0050] Since different data providers may use different label sets, label alignment is crucial to ensure that all participants use the same label set. The label alignment method is the same as step 1.

[0051] The aligned data is encrypted to ensure that the data is not leaked during the calculation process. Specifically, as in step 1, a secret sharing method or a homomorphic encryption method can be used, and the specific secret sharing or homomorphic encryption method also selects the same method as step 1 of this embodiment.

[0052] Step 4: Perform joint training based on dense incremental data to generate an incremental model.

[0053] In this step, the main goal of joint training is to generate an incremental model by processing dense incremental data. This incremental model will supplement and optimize the basic model based on the latest data (such as recently added risk label data). Through joint training, all participants can use incremental data to participate in training together and generate a more accurate incremental model.

[0054] Through joint training and secure aggregation, incremental data will be used to update the global model and generate incremental models. The incremental model is a further optimization of the base model on the incremental data. After the joint training is completed, the performance of the incremental model will be evaluated, and the validation set is usually used to evaluate the accuracy and performance of the incremental model. The performance of the incremental model is compared with the performance of the base model to ensure that the incremental training has brought meaningful optimization.

[0055] Step 5: Combine the basic model and the incremental model to perform model fusion, generate a fusion model to replace the basic model and deploy it to the business system. The goal of this step is to fuse the basic model and the incremental model to generate a new fusion model and deploy it to the business system to replace the original basic model. Through model fusion, the advantages of the two models are utilized to improve the overall prediction effect, so that the system can better adapt to new data features.

[0056] The model fusion method is specifically as follows: Process the incremental data so that its feature distribution is consistent with the basic model input, and generate the prediction results of the basic model and the incremental model; before model fusion, it is necessary to first ensure that the feature distribution of the incremental data is consistent with the input of the basic model. This can be achieved by aligning, standardizing or normalizing the features of the incremental data to ensure that the basic model and the incremental model can output consistent results when processing the same type of input.

[0057] The basic model is used as the teacher model and the incremental model as the student model to train a fusion model.

[0058] Calculate the validation set performance of the basic model and the incremental model, assign weights to the basic model and the incremental model according to the validation set performance, and use the weighted average method to fuse the outputs of the basic model and the incremental model.

[0059] Specifically, during the model fusion process, when training a fusion model, the loss function for:

[0060] In the formula, is the temperature constant used to control the degree of softening, represents the Softmax function, The prediction results of the basic model are is the prediction result of the incremental model, express and The relative entropy of .

[0061] Using the prediction results of the base model and the incremental model, the fusion model is optimized through the above loss function. The fusion model will gradually improve its prediction ability by learning the outputs of the two models, and finally combine the knowledge of the base model and the incremental model.

[0062] In the weighted average method of model fusion, the basic model weights and the incremental model weights are dynamically adjusted through Bayesian optimization.

[0063] Through this fusion process, the final fusion model can combine the advantages of the basic model and the incremental model, continuously optimize in a dynamically changing environment, and demonstrate stronger predictive and adaptable capabilities in actual business.

[0064] In some other embodiments of the present invention, the fusion method of the basic model and the incremental model is expressed as:

[0065] In the formula, Indicates that the fusion obtains a new fusion model. represents the basic model, Indicates an incremental model. It means that based on the prediction results of the existing basic model, the incremental model is optimized through learning brought by new data, and the prediction results of the two are accumulated and fused.

[0066] Or it means that the basic model predicts on the encrypted data of all participants and obtains a sum of predicted values, and the incremental model predicts on the encrypted data of all participants and obtains a sum of predicted values. The final fusion result is the sum of the two. The calculation method is as follows:

[0067] Similarly, in the above formula, Indicates the new fusion model obtained by fusion, N indicates the number of data providers, represents the encrypted label data of user u from the jth data provider, that is, the risk label of user u at data provider j, represents the prediction output of the base model on the encrypted label data, Represents the prediction output of the incremental model on the encrypted labeled data.

[0068] For the above formula and , which is summarized as follows:

[0069] Represents the j-th shard data of the sum of the label vectors of user u, represents the j-th shard of the label vector of user u in participant i, and N is the number of data providers. It means that each participant performs encryption operation on the aligned data set through a secret sharing protocol. The encryption operation is as follows:

[0070] L is the plaintext vector data. After sharding, each participant holds the shard data. , all parties' shard data The sum is L.

[0071] Embodiment 2 This embodiment is a multi-party risk tag real-time data processing system, which is used to implement the real-time data processing method as described in the first embodiment, including: The data acquisition module is used to obtain local user risk label data, continuously collect new user risk label data, and perform label alignment processing to generate an aligned data set; Data encryption module, used to perform secret sharing or homomorphic encryption on the aligned data set; A joint training module, used to obtain encrypted data, and perform joint training based on the encrypted data to generate a basic model; The model deployment module is used to deploy the trained basic model to the business system and apply it in the business system; The incremental training module is used to perform joint training based on dense incremental data to generate incremental models; The model fusion module is used to combine the basic model and the incremental model to generate a fusion model and dynamically adjust the weights according to the validation set performance of the basic model and the incremental model. The model update module is used to replace the basic model with the fusion model and deploy it to the business system.

[0072] The above is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A multi-party risk label real-time data processing method, characterized in that: The following steps are involved: Multiple data providers obtain local user risk label data, and generate aligned data sets through label alignment, and perform secret sharing or homomorphic encryption on the aligned data sets; The task participants obtain the encrypted data, perform joint training based on the encrypted data, generate a basic model, and deploy the trained basic model to the business system; Continuously collect new user risk label data through the business system, perform label alignment on the new data, and then perform confidential processing through secret sharing or homomorphic encryption; Perform joint training based on dense incremental data to generate incremental models; Combine the basic model with the incremental model to perform model fusion, generate a fusion model to replace the basic model and deploy it to the business system.

2. A multi-party risk label real-time data processing method according to claim 1, characterized in that: The label alignment process is specifically as follows: each participant broadcasts its own local label categories to other participants for label alignment, and compares the label categories received by the party with the locally stored label categories, and fills in the missing label categories in the local storage with preset data.

3. The method for real-time data processing of multi-party risk labels according to claim 1 is characterized in that: The secret sharing is specifically: There are N data providers, each of which holds a local user label dataset; For each user label data , construct a random polynomial whose order is less than the set threshold T: In the formula, , indicating that the user On the Participants The The number of tags, is a randomly selected coefficient, T is the minimum number of shares required to reconstruct the data, satisfying ; calculate The value at N different points gives the secret share: Each participant Assign a secret share .

4. The method for real-time data processing of multi-party risk labels according to claim 1 is characterized in that: The homomorphic encryption is specifically: Select a security parameter and have a trusted party generate a public-private key pair , public key Used to encrypt data, private key Used to decrypt data; Each participant Perform homomorphic encryption on local user label data to generate encrypted label data.

5. A multi-party risk label real-time data processing method according to claim 4, characterized in that: The homomorphic encryption adopts one of the Paillier algorithm, the ElGamal algorithm and the Boneh-Goh-Nissim algorithm.

6. The method for real-time data processing of multi-party risk labels according to claim 1 is characterized in that: The model fusion method is specifically as follows: Process the incremental data to make its feature distribution consistent with the basic model input, and generate prediction results for the basic model and incremental model; Use the basic model as the teacher model and the incremental model as the student model to train a fusion model; Calculate the validation set performance of the basic model and the incremental model, assign weights to the basic model and the incremental model according to the validation set performance, and use the weighted average method to fuse the outputs of the basic model and the incremental model.

7. A multi-party risk label real-time data processing method according to claim 6, characterized in that: When training a fusion model, the loss function for: In the formula, is the temperature constant used to control the degree of softening, represents the Softmax function, The prediction results of the basic model are is the prediction result of the incremental model, express and The relative entropy of .

8. The method for real-time data processing of multi-party risk labels according to claim 6 is characterized in that: In the weighted average method of model fusion, the basic model weights and the incremental model weights are dynamically adjusted through Bayesian optimization.

9. The method for real-time data processing of multi-party risk labels according to claim 1 is characterized in that: The joint training adopts a federated learning architecture, where each participant uses local data to train a local model and updates the global model through a secure aggregation mechanism.

10. A multi-party risk label real-time data processing system, characterized in that: The system is used to implement the real-time data processing method according to any one of claims 1 to 9, comprising: The data acquisition module is used to obtain local user risk label data, continuously collect new user risk label data, and perform label alignment processing to generate an aligned data set; Data encryption module, used to perform secret sharing or homomorphic encryption on the aligned data set; A joint training module, used to obtain encrypted data, and perform joint training based on the encrypted data to generate a basic model; The model deployment module is used to deploy the trained basic model to the business system and apply it in the business system; The incremental training module is used to perform joint training based on dense incremental data to generate incremental models; The model fusion module is used to combine the basic model and the incremental model to generate a fusion model and dynamically adjust the weights according to the validation set performance of the basic model and the incremental model. The model update module is used to replace the basic model with the fusion model and deploy it to the business system.

Images